The KERNEL statement specifies the type of kernel and any associated parameters to be used during the semisupervised learning.
You can specify the following kernel-types:
-
KNN </ K=number>
uses a k-nearest-neighbor kernel during training. Specify the number of nearest neighbors as the kernel-parameter, in the form K=number, where number must be a positive integer (the default is 1). For example, specifying KERNEL KNN/K=2 indicates that for each observation, the algorithm finds the two nearest-neighbor samples to calculate the similarity measure, where the distance measure is relying on the Gaussian kernels. The KNN kernel provides a sparse representation of the graph and reduces the memory that is required in order to store the distance matrix for big data.
-
RBF
uses a radial basis function (RBF) kernel during training. No kernel-parameter is needed.
By default, the kernel-type is RBF.